OpenAI CEO Sam Altman said on April 29th that the company was ‘rolling back’ the latest update to the default GPT-4o model powering ChatGPT after users complained about strange behavior — verging on extreme sycophancy. The updates have so far been 100% rolled back, awaiting scheduled fixes.
The AI company published a report on the recent sycophancy issues with the default GPT-4o AI model powering ChatGPT, saying it was rolling back those changes. However, OpenAI emphasized actively testing new fixes to address the issue. It added that it was revising how to collect and incorporate feedback to weigh long-term user satisfaction. The AI firm also disclosed that it was introducing more personalization features that gave users greater control over how ChatGPT behaved.
Users on social media noted that ChatGPT began responding in an overly validating and agreeable way following the GPT-4o model update over the weekend, quickly becoming a meme. They posted screenshots of ChatGPT dishing out all sorts of problematic, dangerous decisions and ideas.
Altman explains what happened and why it matters
the last couple of GPT-4o updates have made the personality too sycophant-y and annoying (even though there are some very good parts of it), and we are working on fixes asap, some today and some this week.
at some point will share our learnings from this, it's been interesting.
— Sam Altman (@sama) April 27, 2025
Altman disclosed that the last couple of GPT-4o updates made the personality too sycophant-y and annoying — even though there were some very good parts of it — and his company was working on prompt fixes, some today and some this week. He also said the company would ‘share its learnings’ eventually.
According to OpenAI, the update — intended to make the model’s default personality feel more intuitive and effective — was informed too much by ‘short-term feedback’ and did not fully account for how users’ interactions with ChatGPT evolved over time.
“We started rolling back the latest update to GPT-4o last night. It’s now 100% rolled back for free users, and we’ll update you again when it’s finished for paid users, hopefully later today.”
~ Sam Altman, CEO at OpenAI
The OpenAI boss clarified that the company was working on additional fixes to model the ChatGPT’s personality and will share more details in the coming days. However, OpenAI explained that the company started with baseline principles and instructions outlined in its ‘Model Spec’ when shaping AI model behavior. It also mentioned that it taught its models how to apply these principles by incorporating user signals like thumbs-up and/or thumbs-down feedback on ChatGPT responses.
OpenAI says users should have more control over how ChatGPT behaves
The AI company believes users should have more control over how ChatGPT behaves and make adjustments — to the extent that it is safe and feasible — if they do not agree with the default behavior. It also said that it was experimenting with ways to let users give ‘real-time feedback’ to directly influence their interactions with ChatGPT and choose from multiple AI personalities.
Particularly, the AI firm said it was exploring new ways to incorporate broader, democratic feedback into ChatGPT’s default behaviors. OpenAI hoped that the feedback would help the company better reflect diverse cultural values worldwide and understand how users would like ChatGPT to evolve.
The AI company also claimed that it was building more safety guardrails to increase its AI’s honesty and transparency, and it would continue to expand its evaluations to help identify issues beyond sycophancy. It added that it was implementing several fixes, including refining its core model training techniques and system prompts to ‘explicitly steer’ GPT-4o away from sycophancy.
OpenAI finally disclosed that it designed ChatGPT’s default personality to reflect its mission and be useful, supportive, and respectful of different values and experiences. However, it acknowledged that each of these desirable qualities, like attempting to be useful or supportive, could have unintended side effects. The AI firm pointed out that a single default could not capture every preference since over 500 million people used ChatGPT weekly across cultures and contexts.
Cryptopolitan Academy: Want to grow your money in 2025? Learn how to do it with DeFi in our upcoming webclass. Save Your Spot
